Edge-based local adaptive thresholding system and methods for foreground detection

Abstract

Systems and methods for generating a locally adaptive threshold image for foreground detection performing operations including creating a saliency edge strength image or layer indicating edge or border pixels of the nuclei by performing tensor voting on pixels neighboring the initial edge pixels within an image region to refine true edges are featured. Further, for each of a plurality of regions or blocks of the image, an adaptive threshold image is determined by sampling a foreground pixel and a background pixel for each initial edge pixel or refined edge pixel, generating histograms for both background and foreground saliency (or gradient magnitude) modulated histograms, determining a threshold range for each block of the image, and interpolating the threshold at each pixel based on the threshold range at each block. Comparing the input image with the resulting locally adaptive threshold image enables extraction of significantly improved foreground.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS[0001]This patent application is a continuation of International Patent Application No. PCT / EP2015 / 064845 filed Jun. 30, 2015, which claims priority to and the benefit of U.S. Provisional Application No. 62 / 018,848, filed Jun. 30, 2014. Each of the above patent applications is incorporated herein by reference as if set forth in its entirety.FIELD OF THE SUBJECT DISCLOSURE[0002]The present subject disclosure relates to digital pathology. More particularly, the present subject disclosure relates to thresholding methods for nucleus segmentation in digital pathology.BACKGROUND OF THE SUBJECT DISCLOSURE[0003]Nucleus segmentation is a key component for many digital pathology applications. A great deal of effort has been made to extract the mask of nuclei or cells from an image comprising of these structures, which is essential for the efficient characterization of these structures. A popular nucleus segmentation framework starts with foreground detec...

Images